24 results about "Relation graph" patented technology

The application relates to the technical field of intelligent reasoning and safety judgment, and discloses a high-speed rail platform safety judgment method and system based on mixed-precision reasoning, which comprises the following steps: acquiring multi-source semantic observation records and generating a safety observation element set; constructing a platform safety relation graph; determining relation conflict density, closed residual error, cross-source divergence degree and reasoning difficulty level; generating a mixed-precision bit width scheduling table; performing graph relation reasoning to obtain an initial safety judgment vector and an initial judgment boundary quantity; in step 6, a final judgment vector is determined; and in step 7, a locked safety level is determined and a safety judgment package is output. The application realizes mixed-precision safety judgment and locked output driven by multi-source semantic relation of a high-speed rail platform.

The application relates to the technical field of video image recognition, and discloses a complex space-time multi-target behavior recognition and understanding system for monitoring video, which comprises the following modules: a video stream analysis module, which acquires frame sequences and determines feature aliasing area data; a feature information extraction module, which extracts human body key points, defines reference geometric nodes and constructs a space-time relation graph; a three-dimensional kinematics constraint module, which determines human body bone length and joint rotation limits as boundary parameters and uses a geometric constraint equation set to obtain the space geometry solution of a hidden point in a three-dimensional space; and a behavior semantic reasoning module, which outputs a behavior semantic label. The application solves the feature aliasing problem caused by target physical interaction by using three-dimensional kinematics rules, deduces a hidden position by using objective geometric constraints, maintains target identity sequence continuity, and improves the reliability of behavior understanding in a complex scene.

This invention discloses a CAN / CAN-FD intrusion detection method based on self-supervised multi-relation graph learning, belonging to the field of vehicle network security technology. The method first slices the vehicle bus message stream into time windows, constructing a multi-relation heterogeneous graph containing temporal and periodic relationship edges for each window, explicitly representing the interaction logic between ECUs and the periodic patterns of messages. Then, a multi-relation graph attention encoder is constructed and pre-trained on unlabeled data using a graph contrastive learning paradigm. Finally, graph-level features are extracted based on the pre-trained encoder and input into a classifier to determine the attack type. This invention does not rely on a large number of labeled attack samples, has the ability to capture complex cross-node interactions, is compatible with both CAN and CAN-FD protocols, and reduces the computational complexity of the inference stage through end-to-end and shallow graph network design. It has the potential for deployment in resource-constrained vehicle bus environments and is suitable for bus security protection scenarios in various intelligent connected vehicles.

The present application relates to a kind of big data platform-oriented stream data incremental cleaning method, belong to data cleaning technical field, comprising the following steps: step 1: data stream is divided into sliding window according to the preset window time length, and adjacent relation graph is constructed according to leaf node co-occurrence relationship;Step 2: the data record in each near neighbor bucket is constructed field fingerprint string, and the data record that field fingerprint string is not identical with all leading mode fingerprints in leading mode fingerprint set is marked as dirty data mode seed;Step 3: with dirty data mode seed as the initial label of dirty data area, the data record of dirty data area is executed local cleaning operation according to original field unit and is recovered after inverse transformation and outputs cleaning result.The present application is applicable to real-time big data scene such as Internet of Things and financial transaction.

The embodiment of the application provides a micro-service topology-aware load prediction method and system based on a graph neural network, which comprises the following steps: obtaining micro-service calling data corresponding to a plurality of micro-service instances in a target container; inputting the micro-service calling data corresponding to a service instance to be predicted into a load prediction model to determine a heterogeneous relation graph based on data construction layers; determining topology features based on a graph structure module, and determining time sequence features based on a time sequence sequence module; performing fusion processing on the topology features and the time sequence features based on a fusion module to output first feature representations; determining candidate micro-service instances according to the first feature representations and the plurality of micro-service instances; performing load state classification processing on the candidate micro-service instances based on a classification layer to obtain load state categories; and determining a target micro-service instance from the candidate micro-service instances based on the load state categories. The technical scheme improves the transparency and credibility of the prediction result, and improves the accuracy and robustness of the micro-service load state prediction.

This invention discloses a method and apparatus for completing a temporal knowledge graph based on a time-series point process. The method includes: classifying nodes in the temporal knowledge graph and obtaining a static entity embedding matrix and a relation embedding matrix; selecting a subgraph in the temporal knowledge graph based on the quadruple to be completed; inputting the subgraph, the static entity embedding matrix, and the relation embedding matrix into a multi-relation graph convolutional neural network to obtain a dynamic entity embedding matrix; performing linear feature modulation on the convolution kernel parameters of the two-dimensional convolutional network based on reference relation embedding to obtain model parameters; and combining the dynamic entity embedding matrix to obtain the confidence score of each node, selecting the node with the highest confidence score as the entity to be completed in the quadruple to be completed. This invention uses linear feature modulation to adjust model parameters, allowing for batch computation of the inference process, resulting in less time consumption when there are many inference targets; the trained model can simultaneously meet the completion requirements for missing entities or relations without requiring separate training.

This invention provides a graph neural network fraud detection method and system based on an attention mechanism. The method includes: calculating the semantic similarity between the central node and its neighboring nodes of a specific relation subgraph in a multi-relation graph based on a label-aware mechanism; adaptively adjusting the selection filtering threshold of neighboring nodes using a reinforcement learning mechanism; aggregating the selected neighboring node information to form a first node representation of each specific relation subgraph in the multi-relation graph; dynamically evaluating the importance of each specific relation subgraph to the central node using an attention mechanism and assigning corresponding weights to form a comprehensive second node representation; fusing the first node representation, the second node representation, and the node representation of the previous layer of the GNN to form a final embedded representation; and inputting the final embedded representation into a classifier to identify and judge fraudulent behavior. This invention achieves efficient detection through sparse gating attention and a neighbor selection mechanism based on reinforcement learning.

This application relates to the field of natural language processing technology and provides a method, apparatus, device, and medium for script asset extraction and index generation. The method includes: splitting the script file into independent texts for each episode, and grouping the independent texts of multiple episodes into several batches of text; in parallel calling a preset large language model to extract structured asset data from each batch of text; if an alias of an asset entity appears as an independent name in a preset directed relation graph, then the preset directed relation graph is updated with the alias of the asset entity as a child node and the standard name of the asset entity as the parent node; based on the updated preset directed relation graph, index records are generated for different names of the same asset entity. This application can automatically and efficiently extract script assets from multi-episode short drama scripts, and can avoid fragmentation and duplicate statistics caused by the same entity appearing with different names in different episodes or batches, forming a unified and searchable asset index.

The application belongs to the technical field of cloth identification, and discloses a cloth material quality identification method and system based on label relations, which comprises the following steps: obtaining a sample set to be measured, semantic features, original classification scores, a label vector of cloth, and constructing an initial relation graph; inputting the initial relation graph into a first graph attention network to obtain shallow relation representation; connecting the semantic features with the shallow relation representation to obtain joint features of each node; inputting the joint features into a second graph attention network to form label encoding; using the label encoding to map the original classification scores into material classification scores corresponding to the sample material quality; inputting the shallow relation representation into a third graph attention network to obtain deep relation representation, and extracting a label compensation vector; using the label compensation vector to compensate the material classification scores to obtain target classification scores, and comparing the target classification scores with positive and negative sample thresholds to determine material information. The application considers the connectivity between different cloths, and significantly improves the identification accuracy of cloth material quality.

The application discloses a kind of indoor multi-scene semantic map construction method based on semantic relation graph, comprising: based on laser radar using grid map algorithm constructs 2D grid map;Using binocular camera obtains scene image, obtains depth map by stereo matching and identifies object in environment using trained target detection model, obtains the class and coordinate of landmark, image data adopts bayesian filtering technology to combine prior information and carries out scene classification, obtains the semantic label of current scene, using the semantic label obtained using the semantic map of occupancy grid method;Based on target identification dataset constructs prior relation graph, fuses space relation knowledge graph and prior relation knowledge graph to construct graph layer, obtains multi-scene semantic map.The semantic relation map of the present application based on deep learning and knowledge graph construction stores a large amount of prior knowledge, gives map comprehensive and accurate environmental information, can realize quick search in multi-scene rescue and daily service.

The application provides a power grid data security protection method and system, relates to the technical field of data security, and comprises the following steps: constructing a data inference relation graph, recording historical access information of an access subject, and when an access request is received, performing two-dimensional risk assessment based on a derived knowledge set inferred by combination of target data and historical data, a correlation measurement value of historical access and an operating state parameter, and generating an access control instruction. The application can effectively prevent inference attacks based on historical data correlation analysis and improve the power grid data security protection capability.

The application discloses an ASD classification feature extraction method based on brain connection enhancement and multi-relation graph aggregation. The feature extraction model used comprises a tangent enhanced Pearson embedding module, a double-channel pooling fusion module, a multi-relation graph construction module and a consensus embedding learning module; an initial individual brain region connection graph is constructed with brain regions as nodes and connection relationships between the brain regions as edges; the tangent enhanced Pearson embedding module constructs a node feature matrix of the initial individual brain region connection graph; the double-channel pooling fusion module extracts and fuses a node feature matrix pair of a fine-grained individual brain region connection graph and a coarse-grained individual brain region connection graph to obtain individual imaging features; the multi-relation graph construction module constructs a non-imaging similarity graph through clustering of non-imaging features and further constructs a multi-relation graph; and the consensus embedding learning module extracts individual ASD classification features according to the multi-relation graph. The method improves the expression ability of brain network features on complex functional activities and the distinguishing ability on ASD.

The application relates to an event prediction method based on pair event relation learning and multi-view evidence fusion, and belongs to the technical field of natural language processing. The method comprises the following steps: pair event relation learning: a pre-training language model is trained by constructing positive and negative samples in a supervised learning framework; an event relation graph is constructed: the pre-training language model after pair event relation learning is used to deduce the dependency relation of the pair event, the probability output by the model is used as the weight of the edge, and thus the event relation graph is constructed; multi-view evidence learning: quantitative evidence of a candidate event is learned from a text semantic view and a graph structure view; credible evidence fusion: the uncertainty of each view is modeled by using a Dirichlet distribution, and multi-view evidence is dynamically fused by using Dempster-Shafer theory to generate a final prediction result. Under the condition of not depending on an external knowledge base, the application realizes a prediction accuracy of 64.62% on an NYT data set, and exceeds an existing baseline model.

The application discloses a kind of multi-target operation object dynamic selection method and system, it is related to computer vision and intelligent video analysis technical field, specifically includes the following steps: S1, obtains the continuous video frame sequence that industrial assembly line monitoring equipment gathers;S2, object detection and tracking are carried out to the video frame, and the object set containing object detection frame, category label and track ID is output;S3, based on the spatial position relationship between object, contact state and function association, construct combination relation graph, identify single body object and the combination target formed by multiple objects.The multi-target operation object dynamic selection method and system, by object combination relationship modeling, multiple object combinations formed operation unit (such as "cover+box", "bolt+nut") can be identified, avoid to combine target error split into multiple independent objects;In automobile parts assembly scene test, the combination target identification accuracy reaches 94.2%.

This invention relates to a knowledge graph completion method in the field of enterprise credit, belonging to the field of artificial intelligence technology. It includes the following steps: S1: data preprocessing and graph structure construction; S2: construction of an efficient multi-relation graph convolutional neural network; S3: dynamic boundary conditions and loss optimization; S4: multi-constraint optimization; S5: training and prediction. This invention significantly improves the completion accuracy of multi-relation knowledge graphs through innovative methods such as multi-relation modeling, dynamic similarity adjustment, and boundary adaptive optimization, resulting in a complete graph structure. It enables the model to maintain high prediction accuracy even with scarce data and low-frequency relationships. Through feature smoothing and cluster consistency regularization, it possesses temporal and dynamic semantic modeling capabilities. It also has the ability to model the "equity chain—supply chain—risk chain" characteristics of the enterprise credit field.

The application discloses a multi-round dialogue recommendation method based on RGCN and reinforcement learning, comprising the following steps: converting user dialogue information into knowledge graph representation, and utilizing a relation graph neural network RGCN to obtain node embedding of the knowledge graph; using multi-head self-attention to obtain good item representation to realize that each item feature embedding is better connected, then using another multi-head self-attention mechanism to capture various user preferences, and using a GRU with an attention update gate to overcome the interference of interest drift; finally, introducing reinforcement learning RL in CRS, which is used for learning a dialogue recommendation strategy, deciding the attributes asked in each dialogue round, the recommended items and when to ask or recommend, and using an improved adversarial Q network DQN to make action decisions, so as to select the optimal next action.

The application relates to the technical field of computers and discloses a semantic scheduling query system and method based on a weak feedback neural algorithm, wherein the semantic scheduling query method based on the weak feedback neural algorithm comprises the following steps: constructing a modal relation graph network, mapping different modal data to a unified semantic space; constructing a condition generation model; and performing weighted integration on a prediction result based on a modal importance evaluation network; constructing a bidirectional mapping self-encoder, realizing knowledge transfer between an existing modal and a newly added modal, and introducing an adversarial verifier to judge whether the converted representation maintains the original semantics; constructing a hierarchical scheduling protocol, performing resource allocation based on query importance and modal resource demand; and dynamically adjusting system parameters by analyzing user implicit feedback signals; the application can provide reliable query services under the condition of modal loss, reduce the adaptation threshold of a new modal, and continuously optimize system performance based on limited implicit feedback signals.

The application discloses an association long-term memory method based on an entity relation graph, comprising the following steps: S1. stable entity and relation extraction; S2. constructing an entity relation graph and upserting to a graph database; S3. dynamic management of relation maximum impression value and edge distance; S4. relation conflict detection and intelligent correction; S5. batch construction of semantic vector data of entities and indexing to corresponding entities of the graph database; S6. triggering of association retrieval; S7. breadth-first association traversal based on BFS; S8. calculation of relation current impression value; S9. sorting and returning of association results. The application can solve the problems in the prior art, such as lack of association reasoning ability, loss of entity relation information, inability to express multiple relations, inability to quantify the importance of relations, lack of time decay mechanism, separation of vector retrieval and graph retrieval, and lack of relation conflict processing mechanism.

The application discloses a tunnel operation and maintenance safety monitoring method and system based on edge computing and a graph neural network, relates to the technical field of tunnel operation and maintenance safety monitoring, and comprises the following steps: an edge processor acquires a unit overall fusion state vector corresponding to each preset short time window of each tunnel unit, and uploads the unit overall fusion state vector and a corresponding space-time identifier to a cloud end; the cloud end constructs a chain space relation graph of the tunnel unit, generates a unit space enhanced state vector corresponding to each preset short time window of each tunnel unit; the cloud end generates a cloud end tunnel unit safety evaluation model; S40, based on the unit enhanced state vector time sequence under the current working condition, prediction is carried out, and an initial safety type evaluation result is generated; the initial safety type evaluation result and neural symbol reasoning are combined to generate a diagnosis result; the method solves the technical problems that existing tunnel operation and maintenance monitoring technology applied to engineering practice is physically untrustworthy and has low monitoring accuracy.

This invention discloses a dual-view drug-disease association prediction method based on a relation graph-guided semantic Transformer. The method acquires drug data, disease data, and known drug-disease association data, constructing a multi-source drug similarity matrix, a multi-source disease similarity matrix, and a drug-disease association matrix. Two complementary heterogeneous graph views are constructed through Top-K local neighborhood sparsification. Node features are initialized based on truncated singular value decomposition features and node degree topological priors. Head-level relation biases are generated using a relation co-occurrence graph encoder and a relation bias generator, and a relation graph-guided Transformer encoder learns drug and disease representations. Further, node-level attention fusion of basic semantics, local semantics, and global semantics, as well as cross-view attention fusion and contrastive learning alignment, yields the final node representation. Finally, drug-disease association scores are output through interaction features and a prediction network. This invention improves the robustness and predictive performance of drug-disease association prediction under sparsity, noise, and class imbalance conditions, and can be used for priority screening of candidate associations for drug relocation.

The application relates to the technical field of knowledge graphs, and discloses a knowledge graph link prediction method fusing multiple sources of information, which uses a graph autoencoder to train a converted relation graph of a knowledge graph, obtains higher-quality entity features and relation features, filters low-correlation neighborhood entities from the semantic and spatial angles, then applies soft hints to the coding of semantic features and hard hints to the coding of high-quality path features, and jointly trains; scores corresponding to a structure modeling model and a text modeling model are preprocessed and feature-extracted, and candidate entities are weighted and summed through weights; when a specific link prediction task is performed, the obtained scores are sorted, and the link prediction task is completed according to the score sorting result. The method alleviates the over-smoothing problem of an existing structure modeling model, reduces the calculation cost, compensates for the entity ambiguity and spatial limitations of an existing text modeling model, and can better complete the knowledge graph link prediction task.

The application discloses a dynamic self-adaptive energy consumption optimization method and device for casting, aiming at the problem that the existing energy consumption management is static and difficult to be optimized online under the constraints of process, safety, beat and maximum demand due to the multi-process coupling and working condition fluctuation of casting, the device set and metering, state acquisition and instruction issuing interface are constructed, the time sequence data is synchronously acquired and the abnormality is marked, the process coupling relation graph of three types of relations is established and the edge weight is dynamically updated, the device state recursive model is constructed, the error boundary is formed by conformal prediction to form the risk margin constraint, the conditional diffusion control instruction generation model is trained, the output is constrained by the behavior support domain, the executable control instruction sequence is generated by combining the control barrier function filtering and multi-step constraint checking, and the execution and rewriting update are issued. The application is used for the full-process collaborative energy consumption optimization control of casting production, reduces the comprehensive energy consumption and peak energy consumption, and improves the operation stability.

The present application relates to the technical field of figure analysis, and particularly relates to a GIS figure entity analysis processing method based on coal mines, which comprises the following steps: obtaining coal mine GIS original figure data, including a plurality of figure entities; preprocessing the original figure data, and extracting a boundary coordinate sequence of each figure entity; calculating a geometric characteristic parameter of each figure entity based on the boundary coordinate sequence, wherein the geometric characteristic parameter comprises an entity length and an entity direction angle; identifying the type of each figure entity according to the geometric characteristic parameter, wherein the type comprises a roadway, a working face or equipment; constructing a topological relation graph among the figure entities based on the type and the geometric characteristic parameter of the figure entities; and outputting the analyzed figure entity data, including the type of the figure entity and the topological relation graph. The present application sets an adjustable judgment threshold, and the system can be flexibly adapted to the figure characteristics of different coal mines, thereby improving the accuracy and automation degree of type identification.

The present application relates to the technical field of geological disaster evaluation, and discloses a ground subsidence and ground fissure disaster evaluation method and system based on a knowledge graph and a graph neural network, comprising the following steps: S1, self-learning ground subsidence and ground fissure coupling danger evaluation knowledge graph construction and rule initialization; S2, knowledge-driven multi-source original data acquisition; S3, evaluation unit construction and unit-level feature matrix generation; S4, original graph structure construction; S5, multi-relation graph structure expansion and feature fusion based on the self-learning knowledge graph; and S6, knowledge-driven combined graph convolution network-based coupling disaster danger evaluation.The present application constructs a self-learning knowledge graph and rule base for ground subsidence and ground fissure coupling danger evaluation, stores domain mechanism knowledge, expert experience and prior constraints in a structured manner, realizes the computability and continuous self-learning update of knowledge, and overcomes the defects of strong subjectivity and static solidification of traditional knowledge-driven models.